A Jittor implementation of Conditional GAN (CGAN).
Introduction
Function of the project
The project build a CGAN which maps a image of random noise and a label to an image of the digital.
The dataset composes of images of different digitals is from MNIST.
Run python CGAN.py at the command line, and the model will begin training. There are several optional arguments, please refer to line 13 to 23 in CGAN.py.
The training will terminate in 100 (as default) epoch. After each epoch, there will be a png image output showing the current effect, which composes of respectively 10 images of digital 0 to 9. After each 10 epoch, arguments of generator will be saved in generator_last.pkl, and arguments of discriminator will be saved in discriminator_last.pkl.
After the training, a result.png image will be generated, which shows the output of mapping with input of a certain number (corresponding to the variable number in CGAN.py).
关于
A Jittor implementation of Conditional GAN (CGAN).
CGAN_jittor
A Jittor implementation of Conditional GAN (CGAN).
Introduction
Function of the project
The project build a CGAN which maps a image of random noise and a label to an image of the digital. The dataset composes of images of different digitals is from
MNIST.How to install the project
The project requires
Jittoras the deep learning framework. Jittor can be installed at https://cg.cs.tsinghua.edu.cn/jittor/download/ with instructions there.How to use the project
python CGAN.pyat the command line, and the model will begin training. There are several optional arguments, please refer to line 13 to 23 in CGAN.py.generator_last.pkl, and arguments of discriminator will be saved indiscriminator_last.pkl.result.pngimage will be generated, which shows the output of mapping with input of a certain number (corresponding to the variablenumberin CGAN.py).